Methods for causal analysis of longitudinal data

点击翻译按钮获取中文摘要

One of the most fundamental questions in medical and health research is: "Does using a particular treatment make sick patients live longer?" In some settings this question can be addressed through the use of a randomized trial however such trials are very expensive and may take many years to complete. An alternative is to try and answer the question using what is known as observational data. That is, data available from completed studies, registries or databases in which the treatment outcome, and a set of patient characteristics have been recorded for a group of subjects. This research program is focused on settings in which treatment is dynamic and changing over time. For example, the dose of a drug that a patient is given over their course of treatment. We assume that subjects receive treatment at regularly spaced intervals (days, weeks, months, etc.) and that as time progresses the level and/or type of treatment they receive is based on their growing history of observed characteristics. A treatment regimen is a rule describing how their next treatment level is determined. Examples of such settings include rules for epoetin dosing to treat anemia in dialysis patients and post-surgical blood transfusion treatment for cardiac patients. This research program is focused on developing statistical models and methods that allow for valid comparison of treatment regimens based on observational data.